Assessment of intra-fractional and inter-fractional motion in esophageal cancer treated with intensity-modulated proton therapy

Abstract

Propose

Proton therapy of esophageal cancer is beneficial to spare normal tissue in clinical practice. However, intra-fractional and inter-fractional variance of tumor motion during treatment may compromise target coverage. The purpose of this study was to investigate the interplay effect due to intra-fractional motion and the effect of the robust optimization parameters for inter-fractional motion in the intensity modulation proton therapy (IMPT).

Materials and methods

This study retrospectively analyzed 42 patients with esophageal cancer treated at Shandong Cancer Hospital. The patients were divided into two groups. Twenty-one patients had a 4DCT image with 10 respiratory phases reconstructed (G_intra). In addition, twenty-one patients underwent a second 3D CT scan following the initial one (G_inter). The RayStation11B treatment planning system was used to create the IMPT plans for these two groups of patients. All patients were planned with a prescribed dose of 50.4 Gy (RBE) in 28 fractions for clinical target volume (CTV). 4D dynamic dose (4DDD) was then calculated to assess the interplay effect by considering respiratory motion and dynamic beam delivery for G_intra. Seven IMPT plans with different robust optimization parameters were designed for the 21 G_inter patient. The setup uncertainties were set to ± 0–6 mm for G_inter. Plan quality and dose-volume histogram (DVH) parameters for the target and organs at risk (OARs) were then analyzed.

Results

For G_intra, 4DDD was slightly perturbated compared to the nominal plan dose. The mean value of CTV D_98% of nominal dose and 4DDD were 49.9 and 49.1 Gy (RBE), respectively, and the CTV D_95% were 50.4 and 49.9 Gy (RBE), respectively. For G_inter, the DVH parameters of the target area and the OARs showed a linear relationship with the corresponding robust optimization parameters in IMPT. When the robust optimization parameters were set with a larger value, the dose coverage of the target was improved. However, the dose of OARs increased at the same time. The D_98% of the target for the seven plans (setup6, setup5, setup4, setup3, setup2, setup1, setup0 plan) were 49.42 ± 0.75, 48.95 ± 1.21, 48.54 ± 1.48, 47.55 ± 2.31, 47.07 ± 2.71, 44.58 ± 4.20 and 44.02 ± 4.44 Gy(RBE), respectively. The D_mean of heart were 12.18 ± 3.05, 11.28 ± 2.86, 10.90 ± 2.77, 9.76 ± 2.39, 9.73 ± 2.39, 8.33 ± 2.22 and 8.22 ± 2.20 Gy (RBE), respectively.

Conclusion

In this study, the differences in dose distributions between the 4DDD and nominal plans for G_intra can be attributed to the interplay effects. While target coverage remained stable, variations in OAR doses should be evaluated for G_intra. For G_inter, smaller setup uncertainty parameters may not fully mitigate inter-fractional tumor motion, leading to greater variation in target dose and potential inadequate coverage. The linear relationship between setup uncertainty and D_98% suggested that improved setup uncertainties can enhance target coverage, while higher setup uncertainties tend to increase OARs doses, particularly to the heart and lungs.

Introduction

Esophageal cancer is a common malignancy of the upper digestive tract with a 5-year survival rate of less than 40% [1, 2]. At present, surgery is the preferred treatment method for early esophageal cancer. However, it is not easy to detect early esophageal cancer, resulting in an inoperable condition for most patients in clinical practice. For patients with advanced esophageal cancer who are not suitable for surgery, chemoradiotherapy is the first choice [3]. However, for esophageal cancer patients with the tumor close to critical organs, it is challenging to balance the delivery of tumoricidal dose and the protection of normal tissue [4].

Proton therapy based on active pencil beam scanning (PBS) has obvious advantages in protecting normal tissue and maintaining dose coverage of the target [5]. Several ongoing clinical trials are investigating its efficacy and safety. For example, NCT01512589 compares the toxicity and progression-free survival between proton beam therapy (PBT) and intensity-modulated radiation therapy (IMRT) to assess the clinical benefits of PBT. NCT05055648 is evaluating the combination of proton therapy and chemotherapy in locally advanced esophageal cancer, aiming to enhance treatment outcomes. Additionally, NCT03801876 is exploring the use of PBT in the neoadjuvant setting, focusing on subsequent surgical outcomes. These studies demonstrate the potential of proton therapy to enhance tumor control and reduce toxicity. However, this indirectly underscores the significant challenge posed by intra- and inter-fraction tumor motion, which remains a key consideration in optimizing proton therapy for esophageal cancer.

Nevertheless, pencil beam scanning proton therapy suffers from greater sensitivity to proton beam range and patient setup uncertainties [6]. Furthermore, due to the esophagus being in close anatomical proximity to the diaphragm, stomach, and other motion organs, the patient’s respiratory movements can cause the esophagus to move. During the intra-fractional treatment process, the motion of the esophageal target, combined with the dynamic delivery of the proton beam, may result in interplay effects. The interplay effect is considered a source of random error. This type of deviation arises from patient-specific random motion or periodic motion such as respiration [7, 8]. Several strategies have been proposed to mitigate the interplay effect, but each has its own limitations. By expanding the margin, range-adapted internal target volume (ITV) could address the respiratory motion [9]. Breath-hold and abdominal compression techniques can reduce the amplitude of motion but decrease patient discomfort during treatments [10]. Breath hold, Gating, and repainting can improve dose distribution at the expense of a longer treatment time [11, 12].

Similarly, inter-fraction motion is the motion seen between images taken on different treatment fractions and has both systematic and random components [13]. The mechanical shortcomings (such as laser misalignment), patient-related factors (like consistent displacement of skin marks) and fixation-related issues may contribute to the errors between inter-fractional treatment [7]. Robust optimization as a way to reduce uncertainty has been widely accepted in clinical practice [14–16]. However, more normal tissue around tumor target will get irradiated as setup and range uncertainties are incorporated in the robust optimization to mitigate the setup and range uncertainties during treatment. Although the development of adaptive proton therapy would be a promising approach to account for the actual errors occur during daily treatment, the use of a suitable robust optimization parameter is beneficial to the robust optimization of esophageal cancer IMPT for the current practice. It can prevent normal tissues from receiving excessive unnecessary irradiation.

The purpose of this study was to investigate the impact of the interplay effect in intra-fractional treatment and assess the target movement between inter-fractional treatment. 33 patients with esophageal cancer treated at Shandong Cancer Hospital were retrospectively analyzed. The patients were divided into two groups. For the intra-fraction group, 4D dynamic dose (4DDD) was calculated to assess the interplay effect by considering respiratory motion and dynamic beam delivery. For the inter-fraction group, IMPT plans with different robust optimization parameters were designed for each patient. The effects of various robust optimization parameters on the alleviation of uncertainty and the protection of OARs were analyzed based on the doses of the target and OARs. This study aims to provide a systematic evaluation for the interplay effect in intra-fractional treatment and the dose variation caused by inter-fractional anatomy for esophageal cancer IMPT.

Materials and methods

Patients and tumor characteristics

This study retrospectively analyzed 42 patients with esophageal cancer treated at Shandong Cancer Hospital. The inclusion criteria were as follows: (1) Patients with unresected esophageal cancer; (2) No previous history of radiotherapy or chemotherapy; (3) No known history of pulmonary or cardiac diseases. The exclusion criteria were as follows: (1) Change of radiotherapy regimen during radiotherapy; (2) Complicated with other malignant tumors.

For each patient, a single CT scan (Brilliance Big Bore, Philips, Netherlands) was performed without breath hold in headfirst supine (HFS) with 3 mm slice thickness. Patients were immobilized using a vacuum cushion (Tengfei Yu Technology Co., Ltd, Shenzhen, China) to ensure reproducibility during CT simulation and treatment delivery. The patients were divided into two groups. Twenty-one patients had a 4D CT image set reconstructed into 10 respiratory phases. The 50% phase was defined as the end of expiratory breathing phase (denoted 0%Ex) which served as the reference phase in this work [17]. Twenty-one patients underwent a second 3D CT scan during the treatment after the first CT scan. The 21 patients were classified as intra-fractional group (G_intra) and 21 patients as inter-fractional group (G_inter).

The gross target volume (GTV), heart, lungs were contoured on all ten phases of 4D CT in the RayStation11B(RaySearch Laboratories AB, Stockholm, Sweden)treatment planning system by the certificated radiation oncologists in our center for this study. The clinical target volume (CTV) was generated by adding a 3 cm longitudinal expansion and 1 cm radial expansion of the GTV on the reference phase [18, 19]. Taking the 50% phase (the end of expiratory breathing phase) as the reference phase, the CTV volume for the G_intra was 130.6–881.9 cc (mean volume 299.5 cc), the CTV volume for the G_inter was 230.2–372.1 cc (mean volume 302.1 cc) during the first scan and 236.9–338.5 cc (mean volume 299.1 cc) during the second scan. Table 1 summarizes patient number, tumor volume, location, motion amplitude (G_intra), and CT interval (G_inter). In this study, tumor motion amplitude was defined as the maximum displacement of the center of the tumor mass in the three coordinate directions throughout all the phases of the 4D CT dataset (S-I, A-P, and R-L).

Table 1.

Patient characterizes for G_intra(patient number, tumor volume, location, motion amplitude) and G_inter(patient number, tumor volume, CT interval)

Patient Group	Gintra						Ginter
Patient Number	Tumor Volume(cc)	Tumor Location	Tumor motion amplitude (mm)			Tumor motion amplitude (mm)	CT interval	Tumor Volume(cc)
			R-L (mm)	I-S (mm)	P-A (mm)
1	357.7	Cervical	0.3	0.5	0.2	0.6	31d	246.6/251.8
2	215.7	Entire esophagus	0.3	1.2	0.1	1.2	28d	356.3/328.7
3	264.9	Entire esophagus	0.1	0.4	0.4	0.6	29d	333.3/310.9
4	196.7	Cervical	0.3	1.5	0.4	1.6	42d	305.3/301.6
5	244.5	Entire esophagus	0.2	0.1	0.1	0.2	21d	311.4/320.7
6	223.7	Thoracic	0.3	2.4	0.3	2.4	33d	230.2/251.6
7	280.5	Entire esophagus	0.6	2.1	0.3	2.2	41d	330.1/301.3
8	881.9	Thoracic	1.0	3.9	1.2	4.2	14d	284.8/306.6
9	417.0	Thoracic	1.2	2.4	1.2	2.9	30d	303.4/307.2
10	130.6	Thoracic	0.2	3.7	1.2	3.9	38d	372.1/336.8
11	365.1	Cervical	0.2	0.6	0.2	0.7	24d	296.0/299.7
12	589.9	Entire esophagus	2.6	4.3	2.0	5.4	15d	287.1/298.6
13	133.76	Lower esophagus	0.8	5.1	1.4	5.3	15d	297.7/285.3
14	401.55	Lower esophagus	1.1	7.6	1.5	7.8	32d	287.8/311.7
15	326.31	Entire esophagus	0.9	1.1	1.2	1.9	35d	319.8/311.0
16	300.65	Lower esophagus	0.6	2.6	0.8	2.7	32d	256.3/236.9
17	245.36	Thoracic	0.5	5.6	0.7	5.7	31d	256.2/283.4
18	145.43	Thoracic	0.8	4.8	2.0	5.3	21d	299.7/300.7
19	144.37	Thoracic	3.3	8.8	3.6	10.1	28d	348.5/338.5
20	238.62	Entire esophagus	0.9	3.2	0.6	3.4	27d	271.2/283.3
21	251.8	Entire esophagus	0.4	1.8	0.7	2.0	27d	350.6/314.5

R-L Right-left, A-P Anterior–posterior, S-I Superior–inferior

Tumor Volume: primary scan time volume/second scan time volume

Treatment planning

The clinical beam model of the ProBeam (Varian Medical Systems, CA, USA) used in the Proton Center of Shandong Cancer Hospital, with the nominal energies of 70–240 MeV in spot scanning mode was implemented. The RayStation11B treatment planning system was used to create the IMPT plans for the two patient groups in this work. The Monte Carlo based algorithm was applied to calculate dose, and the dose grid used was 3 × 3 × 3 mm³. Two oblique posterior fields were adopted [20]. The relative biologic effectiveness (RBE) weighted dose was determined using a fixed RBE value of 1.1 for proton. A prescription dose of 50.4 Gy (RBE) for 28 fractions was administered to 95% volume of the CTV [21, 22]. Only Ginter patients were optimized with robust optimization.

Robust optimization

The plans were optimized using the multi-field optimization (MFO) method. The robust optimization functions were set as follows: The minimum dose of CTV was 50.4 Gy(RBE) with the weight of 100. The maximum dose of CTV was 53 Gy(RBE) with the weight of 100. The dose limits for OARs were as follows: total lung V₂₀ < 25% (V₂₀ defined as the normalized volume of total lung receiving 20 Gy(RBE)), V₅ < 60% (V₅ defined as the normalized volume of total lung receiving 5 Gy(RBE) < 60%), heart dose V₃₀ < 40%, V₄₀ < 30%, maximum spinal cord dose (D_max) < 45 Gy(RBE) [22] and spinal cord (D_max) was treated as D_2% in this work [23].

To evaluate the impact of inter-fractional motion, seven IMPT plans with different robust optimization parameters were designed for each patient in G_inter, and the setup uncertainty was set to ± 0 mm (setup0 plan), ± 1 mm (setup1 plan), ± 2 mm (setup2 plan), ± 3 mm (setup3 plan), ± 4 mm (setup4 plan), ± 5 mm (setup5 plan), ± 6 mm (setup6 plan), range uncertainty was set to ± 3% for all seven plans [24]. In order to reflect the influence of the interplay effect, robust optimization for the IMPT plans of patients in G_intra were not adopted.

Dose calculation

In this study, the interplay effect was evaluated for patients in the G_intra, considering both the dynamic pencil beam delivery and respiratory motion. A proton therapy simulation script [25] was developed in RayStation using the Monte Carlo dose algorithm to calculate the 4D dose distribution (4DDD). The script treatment parameters include: (1) Energy layer switching time: 1 s; (2) Beam spot moving speed: 10 m/s; (3) The beam dose rate was 50,000 clinical-MU/min (approximately 800 MU/s). Total duration of respiratory cycle was assumed to be 5 s. The breathing pattern was represented as a sin curve. All phases were equally distributed in time throughout the respiratory cycle. Additionally, considering the effect of inter-fractional motion, the second CT was registered with the first CT using rigid registration of the bone structures. The proton plan with the first CT was then applied to the second CT set to evaluate the inter-fractional dose variations. A scheme for the process of these two types of dose calculation was shown in Fig. 1.

Fig. 1.

Flowchart for intra-fractional dose, i.e.4DDD calculation and inter-fractional dose calculation

Dose evaluation

The plan dose for each patient in the two groups was normalized so that 100% of the prescription dose covered 95% of the volume of CTV. All patients were evaluated using the same DVH parameters. For CTV, the dose metrics analyzed were D_98% (prescription dose received by at least 98% of volume), D_95%, D_2%, and tumor control probability (TCP). The TCP calculation was based on the equivalent uniform dose (EUD) model. The equations were as follows:

where TCD50 (i.e., the tumor dose required to produce a 50% TCP), γ₅₀ (i.e., the slope of dose response at 50% TCP) and a (i.e., a unitless parameter which describes the magnitude of the volume effect to tumor or the normal structure) were 49.09 Gy(RBE), 0.26, 0.3 [26, 27], respectively. vi is the relative volume related to dose voxel Di. For OARs, total lung V₂₀, V₅, mean heart dose (D_mean), maximum spinal cord dose (D_max). The DVH parameters of the nominal plans and 4DDD were compared to analyze the interplay effect for tumor target and normal tissue. The DVH parameters of the seven plan strategies were linearly fitted to analyze the influence of different setup uncertainty parameters on the dose distribution of target and OARs.

The nominal plans were used as references in the G_intra, and the setup0 plan was used as reference in the G_inter. SPSS software was used for the statistical analysis. The Wilcoxon rank sum test was used to evaluate statistical differences between DVH parameters. A double-tailed p value < 0.05 was considered statistically significant.

Results

Impact of the intra-fractional motion

Figure 2 shows D_98%, D_95%, and D_2% of CTV for the nominal plan and 4DDD of 21 patients in the intra-fraction group. Panel a shows that the nominal plan dose was slightly different from D_98% in 4DDD. Specifically, the D_98% in the 4DDD were marginally lower than the nominal plan dose (exception for one patient). In 4DDD, the mean value of D_98% was 49.9 Gy(RBE). The D_98% was49.1 Gy(RBE) of the nominal plan. Panel b shows that the nominal plan dose was almost the same as that of 4DDD. The mean value of D_95% for nominal plan dose and 4DDD were 50.4 and 49.9 Gy(RBE), respectively. Among the 21 patients, the maximum disparity between the two type of dose was 1.23 Gy(RBE). The median of 4DDD was only 1% lower than nominal plan dose. Panel c shows D_2% of the nominal plan and 4DDD. 4DDDs were mostly higher than the nominal plans. The median of D_2% increased to 100.84% in 4DDD (compared to the nominal plan).

Fig. 2.

The DVH parameters of CTV D_98%, D_95%, and D_2% for the nominal plans and 4DDDs of the intra-fraction group. The red dots represent the nominal plan dose, and the blue dots represent 4DDD

Figure 3 shows the ratios of 4DDD to the nominal plan for D_max in the spinal cord, D_mean in the heart, V₂₀ and V₅ in total lungs. For the majority, 4DDD showed a higher dose for D_mean in the heart. The ratio was mostly above 99% (below 99% in three patients). The D_max of the spinal cord was almost indistinguishable between the two plans, with the ratio above 99% (98.2%, 98.1%in two patients). The performances of V₂₀ and V₅ in total lungs were almost consistent. The difference in V₂₀ was more pronounced than that in V₅ between the two plans. The nominal plan for V₅ differed slightly from the dose for 4DDD, with the ratio concentrated around 100%. In summary, 4DDDs were slightly smaller than the nominal plans for the total lung. The mean values of OARs (spinal cord, heart, V₂₀ and V₅ of total lung were 101.3 ± 2%, 99.62 ± 0.56%, 98.69 ± 1.96%, 100.44 ± 1.38%, respectively.

Fig. 3.

The ratios of 4DDD to the nominal plan for each patient’s OARs in G_intra

For the TCP of the CTV, the TCP in 4DDD was 60.99 ± 0.96%, which was similar to that in the nominal plan (61.22 ± 0.83%). There were statistically significant differences (p<0.05) between the two plans.

Impact of the inter-fractional motion

Figure 4 shows the DVH parameter of CTV in the G_inter after the dose calculation on the second CT image. Panel a represents the values of D_98%. The smaller the setup uncertainty parameter, the dispersed the D_98% values at each point were, and the greater the possibility of insufficient dose coverage of the target. There was a linear relationship between the decrease of setup uncertainty and the decrease of D_98% of target. The fitting equation was

Fig. 4.

Figure 4 This scatter plot shows the DVH parameters D_98%, D_95%, and D_2% of CTV in twenty-one patients of the inter-fractional group. Each dot in the graph represents the patient’s dose of the target under the corresponding setup uncertainty parameter. Each point of the same color represents a single patient. The red line in the figure is a linear fit generated from the mean of these data (dark red is the 95% confidence interval estimate and light red is the 95% prediction interval estimate). The symbols above the dots are the results referring to the statistical significance analysis of the differences of the dose metrics (** p ≤ 0.001; * p ≤ 0.05; ns, p>0.05)

Where y represents the D_98% of the target in unit of Gy(RBE), x represents the setup uncertainties in unit of mm. Panel b represents the values of D_95%. Among the five plans with setup uncertainty of 2–6 mm, the D_98% were more concentrated around 48 Gy(RBE) compared to the other two plans. This phenomenon suggested that when the setup uncertainty was set within the range of 2 to 6 mm, the probability of insufficient dose coverage may decrease. The fitting equation was

Where y represents the D_95% of the target unit of Gy(RBE), x represents the setup uncertainties in unit of mm. Panel c represents the value of D_2%. It was shown that setup uncertainty parameter had little effect on the value of D_2%. In most cases, the doses in the seven treatment plans ranged from 53 Gy(RBE) to 54 Gy(RBE). The fitting equation was.

Where y represents the D_2% of the target in unit of Gy(RBE), x represents the setup uncertainties in unit of mm. There were statistically significant differences (p<0.05) between the six plans and the setup0 plan for the D_98%, D_95%, and D_2% of CTV.

Figure 5 shows the D_mean of the heart, D_max of the spinal cord, V₂₀ and V₅ of total lungs for each case in the inter-fractional group. Panel a represents the average dose received by the heart. The increase of dose was generally consistent with the increase of setup uncertainty parameters. The fitting equation was

Fig. 5.

Figure 5 The DVH parameters of the OARs (D_max of the spinal cord, D_mean for the heart, V₂₀ and V₅ of total lungs) for each case in the inter-fractional group. Each point represents the dose of the OARs with the corresponding setup uncertainty plan. Each point of the same color represents a single patient. The red line in the figure represents the trend of the organ dose with the changing robust parameter settings (obtained by linear fitting of the average organ dose). The dark red is the 95% confidence interval estimate and light red is the 95% prediction interval estimate in the figure. The symbols above the dots are the results referring to the statistical significance analysis of the differences of the dose metrics (** p ≤ 0.001; * p ≤ 0.05; ns, p>0.05)

Where y represents the D_mean of the heart in unit of Gy(RBE), x represents the setup uncertainties in unit of mm. Panel b represents the maximum dose received by the spinal cord. In the twenty-one plans, the dose variance trend of the spinal cord under the seven parameter settings was not uniform for each plan. The dose was mostly concentrated between 40 and 45 Gy (RBE) for the seven setup uncertainties. The dose distribution in the spinal cord seemed to be random and no linear fit was performed. Panel c and d represent V₂₀ and V₅ of total lungs, respectively. The two DVH parameters were affected by setup uncertainty parameters in a consistent trend. The downward trends of the 21 patients were also similar. The fitting equation of V₂₀ was

Where y represents the V₂₀ of the total lung in unit of Gy(RBE), x represents the setup uncertainties in unit of mm. The fitting equation of V₅ was

Where y represents the V₅ of the total lung in unit of Gy(RBE), x represents the setup uncertainties in unit of mm. In general, when the setup uncertainty in the plan settings increase, the radiation dose received by the OARs may also increase. The six plans resulted in significantly (p < 0.05) higher average values than the setup0 plan for the heart, V₂₀ and V₅ of total lungs. However, there was no statistically significant difference (p > 0.05) for the spinal cord, except in the setup1 plan. An increase in the setup uncertainty parameter led to improved dose coverage of the target, resulting in a gradual rise in TCP values. The dosimetric parameters derived from DVH for all treatment plans were summarized in Table 2.

Table 2.

Summary of DVH parameters (mean ± standard deviation) for the seven plans in G_inter

Ginter	D98%	D95%	D2%	TCP	Heart	SC	V20	V5
Setup0 plan	44.02 ± 4.44	46.87 ± 3.00	53.49 ± 0.36	56.80 ± 2.98	8.22 ± 2.20	43.70 ± 2.88	11.90 ± 2.60	21.91 ± 3.80
Setup1 plan	44.58 ± 4.20	47.26 ± 2.65	53.68 ± 0.30	57.36 ± 2.51	8.33 ± 2.22	42.75 ± 5.40	12.20 ± 2.61	22.13 ± 3.61
Setup2 plan	47.07 ± 2.71	48.86 ± 1.53	53.18 ± 0.24	58.47 ± 1.63	9.73 ± 2.39	43.91 ± 2.40	15.10 ± 3.94	24.02 ± 5.43
Setup3 plan	47.55 ± 2.31	49.12 ± 1.27	53.26 ± 0.21	58.73 ± 1.31	9.76 ± 2.39	43.71 ± 2.21	14.89 ± 2.86	24.70 ± 4.23
Setup4 plan	48.54 ± 1.48	49.76 ± 0.94	52.86 ± 0.50	58.84 ± 0.96	10.9 ± 2.77	43.77 ± 1.96	15.71 ± 2.71	27.75 ± 4.52
Setup5 plan	48.95 ± 1.21	49.88 ± 0.53	52.93 ± 0.19	59.12 ± 0.81	11.28 ± 2.86	43.83 ± 1.82	16.28 ± 2.75	28.02 ± 4.71
Setup6 plan	49.42 ± 0.75	50.10 ± 0.29	52.87 ± 0.17	59.34 ± 0.75	12.18 ± 3.05	43.93 ± 1.59	18.08 ± 0.32	30.04 ± 5.67

SC Spinal cord, V₂₀, V₅ the V₂₀, V₅ of total lung

TCP tumor control probability

Units for D_98%, D_95%, D_2%, Heart, and SC were Gy(RBE), TCP, V₂₀ and V₅ were given in percentages (%)

Figure 6 shows the dose distributions for one representative example of G_inter and one of G_intra. For the first part of the figure (G_inter), the dose coverage of the target area gradually decreased as the setup uncertainties were set smaller and smaller. For G_intra, the dose distributions were similar in both panels, 4DDD was sufficient for target.

Fig. 6.

The dose distributions of one representative example of G_inter and one of G_intra. The yellow lines indicated the outlines of the CTVs

Discussion

This study provides a dual-perspective evaluation of both intra- and inter-fractional motion in esophageal cancer proton therapy. A linear assessment of robust parameters under inter-fractional motion was performed. By incorporating TCP analysis, the biological impact of dose variations was further quantified. Among the 21 patients analyzed in the Gintra, the maximum discrepancies between the 4DDD and nominal doses were approximately 1.83 Gy(RBE) for D_98% and 1.23 Gy(RBE) for D_95%. These findings show that the interplay effect of intra-fractional motion has little effect on proton therapy for esophageal cancer in this work. The robust posterior-field distribution used in the IMPT strategies likely explains this result. In terms of anatomical structure, the diaphragm is attached to the lumbar spine via the left and the right crus tendons. This posterior fan-shaped area is relatively stationary with regard to the contracting and relaxing motion of the diaphragm. The posterior-field was not sensitive to changes in respiration and anatomy. Therefore, the plans were more robust [20, 28]. At the same time, the posterior fields may reduce the risk of developing radiation dermatitis [29].

Tumor motion amplitude in this study was defined as the largest motion in three directions (S-I, A-P, or R-L). The maximum motion of the center of mass for the CTV was measured and no baseline markers were implanted in this study. Liu et al. showed that the displacement range in the direction of S-I was the largest, which was consistent with the results measured in this work [30]. Furthermore, Seki et al. reported that tumors located in the lower thoracic esophagus exhibit significantly larger motion amplitudes compared to tumors in the upper or middle thoracic regions (p<0.017) [31]. Tumors in the lower esophagus showed more movement in the S-I and A-P direction than higher tumors [32]. Several studies have shown that tumor location is an important factor in radiotherapy planning for esophageal cancer. In this study, the measured tumor motion amplitudes were consistent with those reported in previous studies. The study by Møller et al. indicated that the respiratory motion depends on the position of the tumor and is largest in the distal part of the esophagus [20]. Similarly, our results showed statistically significant differences in tumor motion amplitudes among different tumor locations (p < 0.05).

Wang et al. have indicated that inter-fractional motion of the tumor occurs during radiation treatments for esophageal cancer, and the volume of the tumor also varies throughout the entire radiotherapy process. Therefore, for inter-fractional motion, the insufficient dose coverage might be caused by the setup errors and the change of tumor volume with the increasing fractions [33]. Our results showed that larger setup uncertainty parameters or adaptive planning is required as the anatomy changes significantly after ~ 20 fractions, which is well align with that reported in Wang’s publication. The use of conformal intensity-modulated radiation therapy (IMRT) or volumetrically modulated arc therapy (VMAT) decreases the normal tissue toxicity. However, the presence of motion may compromise the alignment of steep dose gradients, potentially affecting treatment efficacy [31]. In the process of robust optimization, the setting of robust parameters could ensure the dose coverage of the target area during beam delivery. However, to address setup and range uncertainties during treatment, incorporating these uncertainties into the optimization process may result in increased irradiation of normal tissues surrounding the tumor target. For esophageal cancer patient, in this study, the tumor size of half of patients had reduced. This alteration may result in additional irradiation of normal tissues during inter-fractional treatment. Therefore, it is meaningful to evaluate the robustness parameters in IMPT plans to avoid the excessive setup uncertainty causing more unnecessary damage.

As shown by literatures, the main reasons for inter-fractional changes are gastric filling, deformation of the target volume and changes of the diaphragm position [20]. These changes may result in underdosage of the target for both proton and photon treatments [34–36]. The adoption of posterior beams can reduce the effect of inter-fractional anatomical changes for proton therapy and are more robust to respiratory motion [37]. While the advantages of thoracic proton therapy is controversial [38], for esophageal cancer specifically, robustness of proton plans can be better than IMRT if posterior fields are adopted [39]. In clinical optimization of the IMPT, the physicist can choose the appropriate optimization parameters based on the location and size of the patient’s target area. The range of setup uncertainty used in our study could accommodate the inter-fractional motion amplitude of the majority of patients [40, 41]. As Warren et al. discussed that large setup and range uncertainties > 3.5% could compromise the robustness of the treatment plan [39]. To achieve the goal of CTV D_95% ≥ 95% of the prescribed dose [42], a setup uncertainty of 3 mm is typically required. In addition, for the linear fitting function, its slopes were similar to the dose variance trend for most patients. The intercept varies according to the prescribed dose in IMPT plans. These quantitate results could be useful to balance the target coverage robustness and normal tissue sparing during proton treatment planning. The work published by Lin et al. presented the results of a randomized phase IIB study showing that the reduced doses to OARs translate to reduced toxicity for patients with locally advanced esophageal cancer [43]. Furthermore, the application of the method combining passive scattering and PBS techniques has been shown to translate to a reduction in toxicities for pre-operative patients in a large retrospective analysis [44]. More clinical trials and large retrospective analysis are needed in the era of robust treatment planning.

This study has certain limitations. The small number of cases in G_inter could not fully represent patients with extreme conditions. Patients of G_inter were scanned about 15 or 30 days apart, resulting in limited tumor variation. However, we strongly recommend that once an anatomical change is identified, the radiation therapist and physician team should launch a repositioning CT scan and evaluate whether a new treatment plan is needed. Currently, all cases were analyzed from a single center. In future studies, collaboration with multiple treatment centers could be considered. In this study, two CT images were used to simulate inter-fractional motion. Due to the limitation of the number of CT, the dose variance trend after multiple radiotherapy sessions could not be investigated. The fact that the recalculation has been performed on a second CT over the treatment course may not be sufficient to have a complete assessment of the robustness of the plan. Multiple CTs at different time points would be beneficial. In this study, breathing patterns were assumed to be regular and reproducible based on the 4D CT dataset. In reality, irregular variations in patients’ respiratory motion may occur during the treatment. 4D CT images may not fully capture all aspects of respiratory motion, which may impact the accuracy of 4D dose reconstruction. In subsequent studies, multiple sets of in-room CT images can be introduced for more fractional dose analysis. Future work could include the application of different RBE models to evaluate the effects of RBE uncertainties.

Conclusion

In this study, evaluation of the impact of intra-fractional and inter-fractional motion for a group of 42 esophageal cancer patients receiving IMPT was performed. For G_intra, the observed differences in dose distributions between the 4DDD and nominal plans, particularly in target coverage and OARs, can be attributed to the interplay effects. While target coverage remained largely consistent, the variations in OAR doses should be evaluated for G_intra. Furthermore, the results of G_inter suggest that smaller setup uncertainty parameters may not be sufficient to mitigate the effects of inter-fractional tumor motion, leading to greater variation in the target dose and potentially inadequate target coverage. The relationship between decreased setup uncertainty and reduced D_98% of the target was linear, indicating that better setup precision may improve coverage. For OARs, higher setup uncertainties tended to increase the dose, particularly to the heart and lungs.

Abbreviations

IMPT

Intensity modulated proton therapy

4DDD

4D dynamic dose

PBS

Pencil beam scanning

CT

Computed Tomography

GTV

Gross target volume

CTV

Clinical tumor volume

OARs

Organs at risk

MC

Monte Carlo

RBE

Relative biologic effectiveness

HFS

Head First Supine

R-L

Right-left

A-P

Anterior–posterior

S-I

Superior–inferior

SFO

Single field optimization -

DVH

Dose-volume histogram

MFO

Multi-field optimization

PBT

Proton beam therapy

TCP

Tumor control probability

IMRT

Intensity modulated radiation therapy

VMAT

Volumetrically modulated arc therapy

Authors’ contributions

XYF participated in the design of the study, data collection and the main manuscript writing. STW, WJL and KNM involved in the acquisition of data and revising the manuscript. WH checked and revised the writing parts of the manuscript. YY and TYD participated in the revising of the manuscript critically for important intellectual content. All authors reviewed and approved the final manuscript.

Funding

This work was jointly funded by projects 12105160, 82072094 and 12275162 supported by National Natural Science Foundation of China; projects ZR2019LZL017 and ZR2021QA099 supported by Shandong Provincial Natural Science Foundation; project ts201712098 supported by Taishan Scholars Project of Shandong Province; project 2022B03019-5 supported by Xinjiang Uygur Autonomous Region key research and development project; Shandong Province Higher Education Institutions Youth Innovation Team Development Plan(2024KJJ013); Young Talent of Lifting engineering for Science and Technology in Shandong, China(SDAST2024QTB034) and project 2021HZ81 supported by Tumor precision radiotherapy peak plan.

Data availability

All data included in this study are available upon request by contact with the corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the Research Ethics Board of the Shandong Cancer Hospital. Written informed consent was waived by the Institutional Review Board. This study adhered to the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.